• 

  landing page1

• 

  landing page2

• 

  input page 1

• 

  input page 2

• 

  input page 3

TRUELENS

Technologies Used

API Calls

Value proposition

TrueLens aims to prove advances within the AI space don't inherently come at the cost of human creativity and skills. As fear of replacement continues lingers within fields of Creative Arts, Music, and Computer Science, TrueLens provides a solution that expands access to one of the most impacted industries.

What is TrueLens?

TrueLens is an experimental search agent created by 4 College Students trying to answer the question?

Can AI serve as a tool to expand access to human-created visual content?

Problem:

Throughout the internet, artists and designers were one of the first professionals to experience the fear that 'AI will take my job'. Now as college students, we all constantly hear the discussion around AI surround the belief that it's removing value from our creative output. Sadly, we can clearly agree with a lot of the concerns that arise in this debate. Nevertheless, we still believe AI Agents present a new possibility one to empower artists and help them reach broader markets.

Our hypothesis: Yes!

Our conclusion: It's a little more complicated than we thought, but we are on the right track.

Solution:

TrueLens uses GPT-OSS to process the user's request helping them clearly define the design they are trying to find. Once satisfied it calls a two part search tool:

1. Scraping marketplaces using EXA AI to identify designs currently for sale
2. Vectorizes both the image and the prompt using a finetuned version of OpenClip hosted in Clarifai and returns a similarity score

Potential Changes/Problems we faced

While this MVP doesn't perform as we expected it did help us create a structural framework for the agent, which could be improved upon in the future.

Alternatives for Scraping

1. While Exa AI was incredibly easy to implement, using an finetuned multimodal model capable identifying the right website based on the "image" rather than the "keyword" would present a drastic improvement.
2. Another alternative we would hope to look into is a traditional web scraping API with the GPT-OSS model tailoring the request.

Technologies

GPT-OSS-20B: Serves as the main model processing the user's request.

1. Verifies the request contains all the information necessary for the prompt
2. Calls the Image Search tools when ready or requests additional information
3. The tool scrapes prominent art marketplaces usibg Exa AI and compares the vectorized response images to the prompt to rank them by similarity.

EXA.AI: Identifies websites that match the request and extracts the Image and Site URL

*OpenClip (Clarifai) * Vectorizes both the prompt and response images to ranks images in accordance to similarity with the user's request

Built With

• exaai
• huggingface
• langchain
• openclip
• react
• tailwind